Close Menu
    Trending
    Saturday, September 12, 2026
    Demo

    GPT-6 Astra Users Report Post-Launch Performance Drop After Strong Debut

    Crypto
    GPT-6 Astra Users Report Post-Launch Performance Drop After Strong Debut

    Key points

    • Less than a week after Astra launched, users on X reported noticeable declines in answer quality.
    • Some say weaknesses in writing and reasoning were present from the model’s release.
    • GPT-5.6 Sol faced a similar wave of post-launch criticism in July.

    About a week after its launch, GPT-6 Astra was generating detailed environments and drawing praise for its capabilities. Now, many of those users are sharing screenshots that question the model’s current performance.

    One pseudonymous developer, synthwavedd, wrote on X: “Astra feels significantly dumber for me today.” They added: “Was only a matter of time before The Post-Launch Lobotomy. Shame.”

    Myriad: How high will Nvidia stock go? Click to make your prediction.

    Complaints that AI systems have been deliberately weakened are common after major releases. Perceived performance can change for several reasons, including adjustments to prompts, sampling, tools, or account settings. Nevertheless, the volume and similarity of recent Astra reports have prompted closer scrutiny.

    Developer Pranjal Paliwal, who had previously praised Astra, revisited code the model had written for him and found its quality lacking.

    AGI, or artificial general intelligence, describes a system capable of performing essentially any cognitive task a human can. OpenAI’s president used the term during Astra’s launch, but some users are now applying it to the model’s shortcomings.

    Several complaints describe a similar pattern: faster responses with lower-quality results, along with speculation that OpenAI reduced the model’s “juice value.” One user, Saba, asked why she now has to simplify her prompts to obtain useful answers.

    “Juice value” is not an official OpenAI term. It is generally understood to mean the amount of reasoning or computational effort allocated before a response is generated. Critics suspect that the company reduced that allocation after completing its launch demonstrations.

    A few users attempted controlled comparisons. Salio and researcher Md Ismail Sojal said they submitted the same prompt to Astra at launch and to the current version, receiving weaker results from the latter.

    Other users are reverting to earlier models. Dax Raad, who develops the coding tool Opencode, said his team returned to GPT-5.6 Sol because costs had risen sharply without benefits that justified the change. ChatGPT user Mustafa Sahinli offered a pointed comparison, saying Astra now reminds him of Claude Opus 4.6 “after 1 week of release.”

    Not everyone believes OpenAI intentionally weakened Astra. Pseudonymous user Antikythera offered a detailed counterargument, suggesting that the model’s limitations were evident from the start.

    “It is as dumb as it was on launch,” Antikythera wrote. “The model is good, but the model has a lot of problems. It’s lazy. Writes like a bullet-point-addict… people were overhyped on launch week, now they had time to test it and see its mistakes.”

    Under that interpretation, Astra has not deteriorated; users have simply moved beyond the initial excitement and begun noticing flaws that were present all along.

    Theo, founder of T3Chat, offered a related explanation. He argues that Astra is more variable than Claude Fable, capable of producing exceptional work one moment and poor code the next. With the launch-period enthusiasm fading, more users may now be publicizing its failures.

    OpenAI has encountered a similar pattern before. In July, users reported that GPT-5.6 Sol’s top reasoning mode had become noticeably less rigorous. Executive Tibo Sottiaux denied that the company deliberately weakened it while confirming that OpenAI had been testing controls over reasoning effort—the setting that determines how many steps a model takes before answering.

    One reply reduced the recurring frustration to a joke: a newly released model develops “some kind of disease a few days later” and suddenly becomes less capable. Another suggested that the models may be quantized to reduce companies’ computing costs.

    Quantization reduces the precision of a model’s internal calculations to lower costs, sometimes at the expense of accuracy. OpenAI has not confirmed that it deliberately applied that technique to a shipped model.

    OpenAI has not issued a statement about Astra comparable to its response on Sol. The model remains the company’s first to pass what it calls the critical threshold for cybersecurity risk, meaning it can discover and chain previously unknown software vulnerabilities without human direction. Access to that capability is restricted to vetted defenders through OpenAI’s Daybreak program.

    Regardless of the performance debate, Astra costs $10 per million input tokens and $50 per million output tokens—2.5 times the launch pricing of Sol.

    Also Read

    Source link

    Keep Reading

    New Project (1) (1) (1)
    © 2026 Global Indepth | Developed by Apexify Digital Solutions